Buried in 36Kr's morning roundup is the useful little slap: Zara, H&M and Ralph Lauren were named by Chinese customs. That matters because fashion keeps pretending its biggest operational risks are prompts, personalization and shiny storefront AI. No. The old state machinery still gets a vote, and sometimes it gets the final one.

This is the same structural problem we have been circling in Fashion Runs on AI and Won't Tell You: retail loves invisible complexity right up until a regulator, customs authority or returns auditor drags it into daylight. Pair this with FashionUnited's piece on returns transparency and the pattern is obvious. Brands built black boxes for sourcing, routing and disposition; now governments are starting to pry them open. Good. They should.

Tencent's results coverage, QClaw's full public beta, and Tencent Maps' AI trip planner all point the same way: in China, AI commerce is not being built as a prettier brand website. It is being woven into messaging, mapping, remote control and work software until shopping becomes one more ambient capability inside the stack.

That is why the West keeps misreading China with a homepage brain. Compare this with our own Shopify agents piece and Modern Retail's read on Alibaba. The race is not just agent versus search. It is super-app surface area versus standalone storefronts. One side wants the transaction. The other wants the whole day.

Decathlon entering Pinduoduo looks practical because it is practical. Traffic is hard, customer acquisition is harder, and platforms still know how to put units through the till. But every time a brand says this is just an extra channel, I hear the cash register laughing.

The logic connects straight to agentic commerce and to our March 17 warning in Fashion's Next Channel War. Once discovery, comparison and transaction shift into intermediated environments, the merchant does not just lose margin. It loses sequencing, context and increasingly the right to explain itself before the sale happens. Pinduoduo today, agents tomorrow, same movie.

Modern Retail says Alexa+ users are making three times more purchases on-device, while TechCrunch reports Amazon is working on a smartphone built around Alexa. None of this is subtle. Amazon wants conversation to stop being a layer on commerce and become the container for commerce.

Fashion executives should not dismiss this because the first use cases are batteries, detergent and other joyless things. That is how channel shifts start: with boring replenishment and convenience. Then the interface gets good enough for comparison shopping, gifting, outfit suggestions and replenishable beauty. By the time luxury notices, the customer has already learned a new habit.

This also links neatly to the full catwalk-commerce story we are publishing today without me stepping on its toes: more media surfaces are becoming checkout surfaces, and more checkout surfaces are becoming recommendation engines. The distinction is collapsing.

Creators are pushing back on Instagram's AI-powered Shop the Look test, and they are right to. The platform seems to have learned the classic platform lesson: if people make culture, someone in product will eventually decide that culture is merely unstructured inventory.

Fashion should care because creator imagery is now doing triple duty as marketing asset, training data and commercial interface. That is a grotesque incentive stack. Read this next to our Revolve analysis and you get the broader problem: authorship is being flattened at both ends. AI-generated products muddy creative origin, and AI commerce layers muddy distribution consent. The middleman always says it is improving relevance. Convenient.

Retail Dive cites Gartner saying half of consumers prefer brands that do not use generative AI. On the same beat, Meta is expanding AI enforcement while cutting reliance on human vendors. There is your contradiction in one gulp: platforms are automating harder precisely when users are getting more suspicious of automation.

We already made the broad case in Fashion Runs on AI and Won't Tell You, and there is another trust-focused piece landing today from our side. I will keep this simple: retailers who treat disclosure as optional are not being clever. They are taking a short-term conversion shortcut and booking a longer-term reputation liability. Consumers do not need to understand model architecture to smell evasion.

Cloudflare's CEO says bot traffic could exceed human traffic by 2027, and WordPress.com now lets AI agents publish directly. This is not some distant hygiene issue for ad-tech nerds. It is a commerce measurement crisis waddling toward fashion marketers in broad daylight.

If bots browse products, scrape prices, trigger recommendation systems, generate shopping content and inflate referral patterns, then attribution starts lying long before finance notices. Pair this with our Perplexity-Amazon channel war piece and the message sharpens: brands are heading into a market where access control, traffic quality and measurement integrity all degrade at once. Wonderful. Just what an already over-instrumented industry needed.

This is also one of those moments where two unrelated-looking stories suddenly rhyme: bot pollution online and returns opacity offline are the same managerial disease. Too many retailers are running black-box systems whose outputs they still insist on calling facts.

DoorDash's new Tasks app pays couriers to film activities and submit language data to train AI. I admire the candor, if not the arrangement. At least this version of the industry is no longer pretending that training data floats down from heaven on a cloud.

Retail should pay attention because the same operating instinct is spreading everywhere: turn the edge of the network into a labor-and-data harvesting layer. Couriers, creators, store associates, customer service workers, even shoppers themselves. The glamorous story is AI. The less glamorous story is who gets recruited to feed it, label it, test it and absorb the weird edge cases when it breaks.

This belongs in the same folder as today's labor-politics reporting from our side, even if the sectors differ. Speed, cost pressure and automation are converging on the same answer: somebody lower in the stack gets squeezed and rebranded as efficiency.

Alibaba wants $100 billion in AI and cloud revenue over five years. Fine. Ambition is cheap. Infrastructure is not. The useful way to read this is alongside our recent Alibaba Cloud price-rise report and Fashion Retail's AI Ambitions Are About to Collide With the Compute Bill.

Too many fashion executives still talk about AI like it is a feature bundle you can sprinkle over search, content and clienteling. No. It is becoming a capital allocation question. If China's platforms are integrating agents into super-apps while cloud providers raise prices, then merchants are going to face a brutal tradeoff between buying reach from ecosystems and funding their own intelligence stack. Most will not be able to do both well.

36Kr's Adidas piece reads like a company with strong recent numbers and very little appetite for pretending 2026 will be easy. Sensible. Consumer demand is uneven, channels are fragmenting, and body-shape volatility is no longer a niche planning issue.

The cleanest corroboration is not another earnings call but Destination XL explicitly citing GLP-1-related volatility. We have a full article coming today on why GLP-1 is becoming a merchandising problem, so I will not nick the silverware from my colleagues. But the headline truth is already sitting in plain view: sizing, fit confidence, replenishment and category demand are being nudged by health behavior in ways retailers have barely modeled. Anyone still treating this as lifestyle chatter is asleep.

FashionUnited's interview on Save Our Returns is worth your time because it deals with an unfashionable subject in the literal sense: reverse logistics. This is where retail loves vagueness. Returned goods disappear into a swamp of policies, third parties, write-downs, refurbishment claims and occasional magical thinking.

Connect this to our search piece and to today's customs and bot stories, and a bigger pattern emerges. The industry's data sophistication is wildly uneven. It wants semantic discovery and AI personalization on the front end while still tolerating murk in returns, compliance and post-purchase operations. That asymmetry is not smart. It is childish.

Platum reports on a Korean AI translation platform built to translate entire spreadsheet files for companies going global. Not sexy. Also: very real. Cross-border commerce does not fail because executives lack opinions about agents. It fails because product feeds, care labels, size charts, wholesale sheets and marketplace listings are inconsistent, slow and full of avoidable errors.

This is one reason I keep coming back to structured product data. If agents become the interface, then bad localization stops being a copy problem and becomes a conversion problem. Translation plumbing is not glamorous, but it is closer to revenue than half the demos getting standing ovations right now.